did they actually provide any evidence for this "teaching signal" in the brain?

I'm sure neuroscience community would be very interested in such discovery.

Otherwise this seems like another model looking for the proverbial nail.

Yeah, the problem with models of cognition is that networks are actually a very low-level framework to store information, meaning that both the possible methods to learn and to store are numerous. The model would have to strongly conform to existing data we have of neurology in order to be even worthy of consider to be _the_ cognitive model of the brain.